A new supervised learning approach for visual pattern recognition using discrete circuit elements and memristor array

Abstract

The study of the biology-inspired neuromorphic circuit design is the upcoming field which is the collaboration between neural science and engineering to create a developed circuit that is inspired by the biological world. An efficient system should be designed which is easy to implement and has a remarkable power efficiency and small in size. A bio-inspired neuromorphic circuit system which can be used for pattern recognition has been presented in this paper. A new supervised learning approach technique with a unique control circuit has been developed to train the system to recognize particular patterns which use memristor as synapses and CMOS operational amplifiers as neurons. Each memristor block, which accounts for a single pattern, contains 30 weighted synapses corresponds to one pixel in a 6×5 pixels image. The memristor synapses then weight the voltage pulses that are derived from the pixel values and summed by the output neurons, to be compared with a certain threshold voltage above which an output neuron fires, thus correctly identifying the test image. A system is developed with for 10 patterns (30 memristors corresponding to one pattern) and successful demonstration is shown by training and testing images of numbers from 0 to 9.